Orientation ensemble perception is highly dependent on spatial attention.

The visual world is complex; however, we still access its richness despite computational limits in our visual system. Ensemble perception overcomes this bottleneck by extracting statistics, such as the mean, standard deviation, or other distributional characteristics, of feature sets like orientations or facial expressions, without processing each object individually. The role of attention in ensemble processing is debated: can we extract statistical information about a set when we do not attend to it? To address this question, we used the Posner cueing paradigm in which participants (N = 30, Mean Age = 19.6) reported the orientation of a single Gabor patch or the mean orientation of an ensemble of Gabors. Ensembles consisted of several Gabors whose orientations were sampled from a truncated Gaussian distribution centered on a clockwise or counterclockwise mean. A central arrow pre-cue pointed to either the right or left prior to bilateral stimulus presentation, and a post-cue indicated which side should be reported. In 20% of trials, the post-cued location differed from the pre-cued, so attention was misdirected (invalid trials); in others, cues aligned (valid trials). We used EEG to confirm attentional deployment to the cued location, and we used behavioral measures to estimate the effect of such attentional orientation. We looked for the N2pc, an EEG marker of lateralized attention, but could only detect it in the single Gabor condition. However, multivariate pattern analysis (MVPA) confirmed attentional shifts by showing above-chance classification accuracy of the attended side in both single and ensemble conditions. Behavioral performance was measured using d', a measure of sensitivity. Performance was impaired on invalid compared to valid trials in both the single Gabor and ensemble conditions, consistent with a role of attention in both cases. Accuracy dropped to chance levels on invalid ensemble trials but remained above chance, though reduced, on single trials. Overall, these findings indicate that misdirecting spatial attention severely impairs the ability to report the mean orientation of peripherally presented ensembles.